The first automation matters more than the ones after it. Pick well and everybody wants the next one. Pick badly and you've spent two weeks proving to your team that this was a distraction.
So here's how to choose, and the arithmetic that settles most arguments about it.
Frequency beats size
The instinct is to pick the job everybody complains about, which is usually the big one. The monthly report. The quarterly stocktake. The annual renewal run.
That's the wrong end. A three-minute job done twenty times a week is fifty hours a year. A three-hour job done monthly is thirty-six. The small one is worth more, and it's easier to build, easier to test, and it proves itself in a couple of weeks instead of at the end of the quarter.
There's a second reason and it's the one that matters more. A job that runs twenty times a week gets its bugs found in the first couple of weeks. A job that runs monthly hides them until March.
The arithmetic
Take each candidate and work out three numbers.
- Hours a year. Times per week, times minutes each, times fifty, divided by sixty.
- What an hour is worth. Not somebody's wage. What they'd be doing instead, which for most small businesses is work you can bill for or work that brings in customers.
- What it costs to set up. Money if you're paying somebody, and honestly counted hours if you're doing it yourself.
If the first two multiplied together beat the third inside a year, it's worth doing. Most genuinely good candidates clear that in three or four months and it isn't close.
There's a saving that isn't time. An appointment reminder saves two minutes of texting and prevents a no-show, and the no-show was a whole empty slot. Count that separately and count it honestly.
Four questions before you commit
- Does it run the same way every time? If there are five exceptions, you've got five jobs, and the automation only covers one of them.
- Can two people agree on the steps? If they can't, that disagreement is the actual problem and no software fixes it.
- Would somebody be relieved to lose it? They'll help you get it right and tell you when it's wrong. An automation nobody asked for gets worked around within a month.
- Is a mistake cheap and obvious? Start where a wrong answer costs a minute. Save the money and the legal obligations for your third project, not your first.
The candidates that look tempting and aren't
Three that come up constantly and are worth resisting at the start.
Anything where the exception is the point. If every job needs a judgement call, that isn't a process. Forcing it into one costs you the flexibility your customers are paying for.
Whatever your competitor announced. You have no idea whether it works for them, and you're picking a project from a press release instead of from your own week.
Quoting. It's usually the biggest prize and it's a poor first project, because a wrong quote costs real money and the arithmetic is more tangled than it looks. Do it third, once the team trusts the first two.
How to run the first one
Set a date to check it, before you build it. Two weeks out.
Have it run alongside the person doing the job for the first week, not instead of them. You're comparing two answers, and that's the only way to find out how often it's wrong before it matters.
Then measure the same thing you measured at the start. If nobody can tell you what changed, it didn't work, and the useful move is to say so and pick a different job. That's a cheap lesson if you took the small one first.